Mismatch vs Bottleneck
Two failure modes in agent setups - one is dangerous, one is just wasteful.
What is a mismatch vs a bottleneck in AI agent setups?
In AI agent setups, a mismatch occurs when your Agent Access Level exceeds your Agent Power Level (AAL > APL) - your agent can reach more than you can control. This is a critical structural risk. A bottleneck occurs when your Power Level exceeds your Access Level (APL > AAL) - you have powerful tools but limited access, wasting potential. The goal is alignment: your control should match or exceed your agent's reach.
How do APL and AAL create risk when misaligned?
Your agent setup has two coordinates: what your tools can do (APL) and what they can reach (AAL). Problems occur when these do not match.
What is a mismatch (AAL > APL)?
A mismatch happens when AAL > APL. Your agent has access to systems, files, or data that you cannot adequately govern.
Examples
- Using a vibe-coding tool (APL2) with write access to your entire codebase (AAL 3)
- Running a multi-agent pipeline (AAL 4) without understanding how to monitor or stop it (APL2)
- Granting production database access to an assistant because it was convenient
Why is a mismatch dangerous?
You cannot verify what you cannot understand. If your agent has access beyond your control level, you are trusting it blindly in areas where you cannot catch its mistakes.
The AI Setup Snapshot flags this as CRITICAL risk.
What is a bottleneck (APL > AAL)?
A bottleneck happens when APL > AAL. You have powerful tools but they are not connected to the systems where you could use them.
Examples
- You know how to run Claude Code (APL3) but only use it for chat (AAL 1)
- You have built multi-agent frameworks (APL4) but only use them for single-step tasks (AAL 2)
- You have sophisticated workflows but they only touch test environments
Why is a bottleneck wasteful?
You've invested in capabilities you're not using. This isn't dangerous - it's just inefficient. The Snapshot flags this as STAGNANT.
What is the healthy state for an AI agent setup?
The goal is alignment: APL ≥ AAL. Your control matches or exceeds your access. You can verify, understand, and govern everything your agent tools can reach.
How do you fix a mismatch?
Two options:
- Reduce access (AAL) - Restrict what the agent can reach until it matches your control level
- Increase control (APL) - Learn to operate at a higher level before granting more access
Option 1 is faster. Option 2 is more powerful long-term.
How do you fix a bottleneck?
Gradually increase access while maintaining governance:
- Start with low-stakes systems
- Build verification workflows before expanding access
- Move from test environments to staging to production